Adapting Foundation Models for Annotation-Efficient Adnexal Mass Segmentation in Cine Images

Francesca Fati; Alberto Rota; Adriana V. Gregory; Anna Catozzo; Maria C. Giuliano; Mrinal Dhar; Luigi De Vitis; Annie T. Packard

cs.CVApr 9, 2026

Adapting Foundation Models for Annotation-Efficient Adnexal Mass Segmentation in Cine Images

Authors: Francesca Fati, Alberto Rota, Adriana V. Gregory, Anna Catozzo, Maria C. Giuliano, Mrinal Dhar, Luigi De Vitis, Annie T. Packard,

Organizations: Mayo Clinic · Politecnico di Milano · Istituto Europeo di Oncologia

Abstract

Adnexal mass evaluation via ultrasound is a challenging clinical task, often hindered by subjective interpretation and significant inter-observer variability. While automated segmentation is a foundational step for quantitative risk assessment, traditional fully supervised convolutional architectures frequently require large amounts of pixel-level annotations and struggle with domain shifts common in medical imaging. In this work, we propose a label-efficient segmentation framework that leverages the robust semantic priors of a pretrained DINOv3 foundational vision transformer backbone. By integrating this backbone with a Dense Prediction Transformer (DPT)-style decoder, our model hierarchically reassembles multi-scale features to combine global semantic representations with fine-grained spatial details. Evaluated on a clinical dataset of 7,777 annotated frames from 112 patients, our method achieves state-of-the-art performance compared to established fully supervised baselines, including U-Net, U-Net++, DeepLabV3, and MAnet. Specifically, we obtain a Dice score of 0.945 and improved boundary adherence, reducing the 95th-percentile Hausdorff Distance by 11.4% relative to the strongest convolutional baseline. Furthermore, we conduct an extensive efficiency analysis demonstrating that our DINOv3-based approach retains significantly higher performance under data starvation regimes, maintaining strong results even when trained on only 25% of the data. These results suggest that leveraging large-scale self-supervised foundations provides a promising and data-efficient solution for medical image segmentation in data-constrained clinical environments. Project Repository: https://github.com/FrancescaFati/MESA

Figures & tables

Figure 2: Model Architecture Overview : The input image $\mathbf{I}$ is processed by the DINOv3 Encoder $E$ which extracts a set of hierarchical feature maps from which we retain only the subset $\{\mathbf{F}_{\ell_{0}},\mathbf{F}_{\ell_{1}},\mathbf{F}_{\ell_{2}},\mathbf{F}_{\ell_{3}}\}$ . The feature maps are then passed to a learned resampling operator $\psi$ that resizes them to a higher or lower resolution depending on their rank in the hierarchy, $\{\mathbf{G}_{\ell_{0}},\mathbf{G}_{\ell_{1}},\mathbf{G}_{\ell_{2}},\mathbf{G}_{\ell_{3}}\}$ . Later, the Upsample and reassemble modules $\phi$ progressively fuse the resampled maps into a final representation that is processed by the Segmentation head $H$ , to obtain the final segmentation mask $\mathbf{S}$ .

Figure 3: Qualitative Segmentation Comparison : Visual comparison of adnexal mass segmentation results across different architectures. Our method demonstrates superior boundary fidelity and structural consistency.

Method	Dice $\uparrow$		IoU $\uparrow$		Sens. $\uparrow$		HD95 $\downarrow$		MSD $\downarrow$
Method	224	512	224	512	224	512	224	512	224	512
U-Net	0.939	0.880	0.884	0.785	0.942	0.833	11.13	65.886	2.664	20.405
U-Net++	0.935	0.875	0.878	0.778	0.937	0.823	10	65	2.703	20.143
DeepLabV3	0.916	0.932	0.845	0.872	0.934	0.949	12	18.868	3.526	5.842
MAnet	0.924	0.932	0.859	0.873	0.941	0.953	10.851	21.201	3.18	6.378
OURS	0.945	0.937	0.897	0.882	0.956	0.956	8.861	18.455	2.288	5.323

Table 1: Quantitative comparison of segmentation performance of adnexal mass segmentation performance across the related work at resolutions

224\times 224

and

512\times 512

. For ’

\uparrow

’, higher is better, for ’

\downarrow

’ lower is better: bold typeface indicates the best performing method.

Figure 4: Efficiency Analysis Results : a) Pareto efficiency analysis comparing segmentation performance against model capacity. We fit a logarithmic curve to the methods based on DINOv3 and to the convolutional state of the art separately and we report the log slope $a$ and the fit quality $R^{2}$ . b) Data Starvation Curves report the performance losses at progressively larger starvation fractions.

	Retention Index $RI$
Method	Dice	IoU	Sens	HD95	MSD
UNet	0.884	0.803	0.808	0.381	0.300
UNet++	0.897	0.823	0.841	0.400	0.376
DeepLabv3	0.866	0.779	0.749	0.303	0.353
MAnet	0.897	0.824	0.843	0.429	0.458
DINOv3-small	0.970	0.944	0.954	0.802	0.800

Table 2: Retention Indexes for all metrics under -75% data starvation. Best performance is in bold typeface

Figure 5: Convergence efficiency across architectures. Area under the Learning Curve (ALC) for different models and backbone sizes at two input resolutions. Each pair of markers shows the same architecture at different resolutions.

Medical Image Segmentation

Medical image segmentation delineates organs, lesions, and other structures in scans to support measurement, diagnosis, treatment planning, and research.

See all 788 papers

SegDINO: Introducing Multi-Scale Structure into DINO for Efficient Medical Image Segmentation
Jun 16, 2026Sicheng Yang, Hongqiu Wang, Zhaohu Xing +5Medical Image Segmentation
The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China · Department of Radiology, State Key Laboratory of Oncology in South China, Guangdong Provincial Clinical Research Center for Cancer, Sun Yat-sen University Cancer Center, Guangzhou 510060, China · Department of Pancreatobiliary Surgery, State Key Laboratory of Oncology in South China, Sun Yat-sen University Cancer Center, Guangzhou, Guangdong, China +1
DINO-MVR: Multi-View Readout of Frozen DINOv3 for Annotation-Efficient Medical Segmentation
May 8, 2026Wei Jiang, Feng Liu, Nan Ye +1Image Segmentation Medical Imaging Foundation Models
The University of Queensland · The University of Newcastle
SemiSAM-O1: How far can we push the boundary of annotation-efficient medical image segmentation?
Apr 27, 2026Yichi Zhang, Le Xue, Bichun Xu +6Image Segmentation Medical Imaging Foundation Models
Artificial Intelligence Innovation and Incubation Institute, Fudan University, Shanghai, China. · Shanghai Academy of Artificial Intelligence for Science, Shanghai, China. · Department of Radiotherapy, Tongji Hospital, School of Medicine, Tongji University, Shanghai, China. +2

Jun 16, 2026J/K move · Enter open · S save

SegDINO: Introducing Multi-Scale Structure into DINO for Efficient Medical Image Segmentation

Sicheng Yang, Hongqiu Wang, Zhaohu Xing, Sixiang Chen+4

The Hong Kong University of Science and Technology (Guangzhou), Guangzhou, China · Department of Radiology, State Key Laboratory of Oncology in South China, Guangdong Provincial Clinical Research Center for Cancer, Sun Yat-sen University Cancer Center, Guangzhou 510060, China · Department of Pancreatobiliary Surgery, State Key Laboratory of Oncology in South China, Sun Yat-sen University Cancer Center, Guangzhou, Guangdong, China+1

Medical Image Segmentation

Open paper AskPDF arXiv

Self-supervised DINO models provide strong transferable visual representations, yet applying them directly to image segmentation remains challenging. Existing approaches commonly rely on heavy decoders with complex upsampling, introducing substantial parameter and computational overhead. We observe that introducing scale into DINO features is far more critical than increasing decoder capacity. In this work, we present SegDINO, an efficient segmentation framework that integrates a DINOv3 backbone with lightweight scale modeling. SegDINO introduces Token Pyramid Adaptation (TPA) to reorganize intermediate DINO features into a pseudo multi-scale hierarchy, and Scale-Aware Decoding (SAD) for efficient intra-scale refinement and top-down multi-scale propagation. We further curate PanCT, a new CT dataset containing 284 patients with expert-annotated pancreatic tumors, to assess SegDINO's ability to handle difficult small-lesion cases. Extensive experiments on PanCT and three public benchmarks demonstrate that SegDINO achieves state-of-the-art results with high efficiency. The code is available at https://github.com/script-Yang/segdino_v2.

Abstract

Medical Image Segmentation

Explore similar work

SegDINO: Introducing Multi-Scale Structure into DINO for Efficient Medical Image Segmentation

DINO-MVR: Multi-View Readout of Frozen DINOv3 for Annotation-Efficient Medical Segmentation

SemiSAM-O1: How far can we push the boundary of annotation-efficient medical image segmentation?